Enhancing Disease Classification in Chest X-ray Images: A Comparative Study of Preprocessing Techniques for Convolutional Neural Networks

In this study, we propose a Convolutional Neural Networks (CNN) model for classifying diseases using chest X-ray images. Based on a dataset comprising images classified into COVID-19, pneumonia, and normal lungs, provided by Kaggle, we extracted features using three different methods. The first method involved basic preprocessing such as resizing images and changing grayscale. The second method incorporated Histogram of Oriented Gradients (HOG) in addition to basic preprocessing, while the third method added Contrast Limited Adaptive Histogram Equalization (CLAHE) to the basic preprocessing. Despite conducting comparative experiments to enhance model performance, all three methods yielded similar accuracies ranging from 0.94 to 0.95. However, each CNN model trained on preprocessed images effectively classified the test dataset. The base model of each preprocessed model maintained excellent performance regardless of the preprocessing technique used. The proposed model in the paper has 141,827 parameters, fewer than mobilenet, designed for lightweight models, offering the advantage of deploying deep learning models even in memoryconstrained environments. Therefore, this study emphasizes the potential for further enhancing model performance by applying better preprocessing techniques, and constructing applications or programs to be used in clinical settings to contribute to the early detection or accurate diagnosis of diseases in patients.